The impact in software has been very hard to measure. There are so many ups and downs and variables.

Adoption in particular is a useless metric. They are forced to adopt even if it's not really helping in their case, or if it does help but using it makes them miserable, like being forced to switch jobs from something you enjoy to something you find boring and tedious. And then there's the "expertise debt" that will have who knows what impact in the coming decades.

Also, this "Legacy Code as a Service" tech debt boom may eventually find creditors that need paying. These models are trained on other people's Legacy Code. No company is letting OpenAI or Anthropic or Meta have access to their best code and smartest experts. It's all just other people's Legacy Code. LLMs are going to force companies towards a lot of mediocre software regression to the mean. How long before companies are scrambling for expertise they laid off or didn't invest in because all their software sucks and no one around knows what it is doing or why? The tech debt bank account is going to get weird in the coming decades.